The model and the app are different things
When you use a chatbot, you are really using two things at once. One is the model, the trained language model that predicts text. The other is the app: the website, phone app, or program you type into. The app takes your message, sends it to the model, and shows you the reply.
A handy comparison is a car engine and the car around it. The engine does the core work, but the rest of the car decides what you can do with it, from the seats to the GPS. Two cars can share the same engine and still feel very different to drive.
The same model can sit inside many apps. A company might offer it in its own chatbot, while other companies build it into email programs, writing tools, or customer service chats.
What apps can add
On its own, a model can only read what's in its context window and write text back. Apps often give it extra abilities, usually called tools. Common ones include:
- Web search. The app looks things up and passes the results to the model, so it can talk about recent events.
- File reading. You upload a PDF or spreadsheet, and the app gets it ready for the model to read.
- Memory. The app saves notes about you and adds them to future chats.
- Image making. The app creates pictures, often by handing your request to a separate image model.
- Running code. The app runs small programs to do exact math or sort data, instead of the model guessing at numbers.
In many apps, the model can decide when to use a tool. It might say, in effect, "search for this," and the app does the search and hands back the results. The model then writes its answer using them.
Why the same model acts differently
Because of all this, one model can behave differently from one product to the next. An app might add hidden instructions before your message, like "You are a polite helper for a bank. Keep answers short." This is often called a system prompt, and you usually can't see it. One app might have search turned on and another not. One might have memory, while another starts fresh every time.
So if a chatbot at work answers differently than the one you use at home, it may not be a better or worse model. It may just be a different setup around it.
Try it
Open a chatbot you use and look at its settings and the buttons near the message box. Jot down the tools it offers, such as search, file upload, or memory. Then ask it, "What tools can you use in this chat?" and compare its answer to what you found.